A model registration method and system based on oral image three-dimensional reconstruction

Abstract

The application provides a model registration method and system based on oral cavity image three-dimensional reconstruction, the method comprises the following steps: obtaining an initial three-dimensional model based on a preset AI segmentation algorithm and pre-acquired oral cavity CBCT data; obtaining a triangular mesh model based on a preset three-dimensional reconstruction algorithm and the initial three-dimensional model; obtaining a target three-dimensional model based on a preset model simplification algorithm and the triangular mesh model; obtaining a physical space center based on the pre-acquired oral cavity CBCT data, and registering the image center of the target three-dimensional model with the physical space center based on a preset model space registration algorithm. The application solves the problem that the three-dimensional model is easy to lose space information and physical information, resulting in low model accuracy, after the target file is extracted by the automatic segmentation algorithm in the prior art. The application realizes three-dimensional model reconstruction simplification and space registration, ensures the structure information and structure positioning accuracy of the model, and improves the model accuracy.

Description

Technical Field

[0001] This invention relates to the field of digital healthcare, and in particular to a model registration method and system based on three-dimensional reconstruction of oral images. Background Technology

[0002] Cone-beam computed tomography (CBCT) is a commonly used imaging technique in oral medicine, reflecting oral anatomical information such as teeth, maxillary sinuses, and mandibular canals. Target structures can be accurately extracted from CBCT images and standardized three-dimensional models can be generated, transforming a two-dimensional problem into a three-dimensional one. This allows for more precise analysis of the information contained in CBCT images. Therefore, how to more accurately obtain three-dimensional models has become a key research issue.

[0003] Currently, existing technologies extract target files through automatic segmentation algorithms and then convert them into 3D models in formats such as STL. However, existing technologies, such as AI algorithms, rely on automatic segmentation, which can easily lead to the loss of spatial and physical information in the 3D model after the target file is extracted. Therefore, existing technologies have the problem of low accuracy in 3D model construction. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a model registration method and system based on three-dimensional reconstruction of oral images. This method simplifies the reconstruction of the three-dimensional model and ensures accurate structural information and positioning, thereby improving model precision.

[0005] To achieve the above objectives, embodiments of the present invention provide a model registration method based on three-dimensional reconstruction of oral images, comprising: obtaining an initial three-dimensional model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data; obtaining a triangular mesh model based on a preset three-dimensional reconstruction algorithm and the initial three-dimensional model; obtaining a target three-dimensional model based on a preset model simplification algorithm and the triangular mesh model; obtaining the physical spatial center based on the pre-acquired oral CBCT data; and registering the image center and physical spatial center of the target three-dimensional model based on a preset model spatial registration algorithm.

[0006] This invention proposes a model registration method based on 3D reconstruction of oral images. After processing pre-acquired oral CBCT data using a preset AI segmentation algorithm to obtain an initial 3D model, a triangular network model is first constructed using the preset 3D model to perform 3D reconstruction. Then, the reconstructed triangular network model is simplified using a preset model simplification algorithm to obtain the target 3D model, thereby improving the geometric accuracy of the model. Finally, the physical spatial center is obtained from the pre-acquired oral CBCT data, and the physical spatial center is registered with the image center of the target 3D model according to a preset model spatial registration algorithm, ensuring that the model and the original image are aligned in the physical coordinate system. Thus, by employing steps such as AI segmentation, 3D reconstruction, model simplification, and spatial registration to simplify and spatially register the 3D model, the structural information and structural positioning of the model are ensured to be complete and accurate, improving the model's precision.

[0007] Furthermore, based on the preset AI segmentation algorithm and the pre-acquired oral CBCT data, an initial three-dimensional model is obtained, including: obtaining several slice files based on the pre-acquired oral CBCT data; sorting the slice files according to their file numbers to obtain the slice file paths; traversing each slice file in the slice file path to obtain the number of slices and the physical information of each slice file, and obtaining spatial structure information based on the number of slices and the physical information of each slice file; and obtaining the initial three-dimensional model based on the preset AI segmentation algorithm and the spatial structure information.

[0008] In the above scheme, the slice files corresponding to the pre-acquired oral CBCT data are sorted, and the slice file paths are constructed based on the sorting results. This ensures that the pre-acquired oral CBCT data maintains the correct spatial order and structural continuity during 3D reconstruction. Then, the number of slices and physical information are extracted to construct accurate spatial structural information, providing an accurate geometric basis. Finally, an initial 3D model is constructed using a preset AI segmentation algorithm. Therefore, accurate analysis of the pre-acquired oral CBCT data provides an accurate data foundation for subsequent 3D reconstruction, model simplification, and spatial registration steps, ensuring the integrity and accuracy of the model's structural information and location, and improving the model's precision.

[0009] Furthermore, based on a preset AI segmentation algorithm and spatial structure information, an initial 3D model is obtained, including: automatically segmenting the pre-acquired oral CBCT data using a preset AI segmentation algorithm to obtain a first oral mask image; traversing several voxels in the first oral mask image, identifying corner voxels and setting the corner values ​​to preset foreground values ​​to obtain a second oral mask image; constructing the second oral mask image into a visual image object, and based on the spatial structure information, setting the spatial dimension and actual physical size of the visual image object to obtain the initial 3D model.

[0010] In the above scheme, based on the AI ​​segmentation algorithm, the pre-acquired oral CBCT data is automatically segmented to obtain a first oral mask image. Then, foreground values ​​are assigned to the corner voxels of the first oral mask image to obtain a second oral mask image. The processed second oral mask image is converted into a visual image object with spatial dimensions and actual physical space size, ensuring that the initial 3D model maintains structural integrity. Thus, the 2D image is converted into a 3D model. Through mask assignment, not only is the structural integrity of the 3D model guaranteed, but the 3D model also has physical spatial attributes consistent with the actual image, ensuring the integrity and accuracy of the model's structural information and structural positioning, thereby improving the model's accuracy.

[0011] Furthermore, based on the preset 3D reconstruction algorithm and the initial 3D model, a triangular mesh model is obtained, including: constructing several polyhedral units from adjacent voxels in the initial 3D model based on the preset 3D reconstruction algorithm; extracting boundary surfaces from several polyhedra based on preset isosurfaces; traversing several polyhedra and generating triangular patches based on the boundary surfaces; and meshing the triangular patches to obtain the triangular mesh model.

[0012] In the above scheme, voxels in the initial 3D model are constructed into polyhedral units, and boundary surfaces are extracted based on preset isosurfaces. Triangular patches are then generated from these boundary surfaces, and after mesh stitching, a triangular mesh model is obtained. This allows a preset 3D reconstruction algorithm to accurately capture the contours of the complex anatomical structures of the initial 3D model, generating a continuous and closed triangular mesh model. Therefore, by converting volume data into a triangular mesh, the accuracy of 3D reconstruction is ensured while also improving visualization effects. This guarantees the integrity and accuracy of the model's structural information and positioning, thus enhancing the model's precision.

[0013] Furthermore, based on a preset isosurface, boundary surfaces are extracted from several polyhedra, including: mapping several voxels in the initial 3D model to binary masks; dividing voxels that do not meet the preset mask requirements into a first voxel group according to the preset isosurface; dividing voxels that meet the preset mask requirements into a second voxel group; and extracting boundary surfaces from several polyhedra based on the first voxel group and the second voxel group.

[0014] In the above scheme, voxels are mapped to binary masks, and voxels are grouped according to isosurfaces. The target region is distinguished from the background region by the first and second voxel groups to extract the boundary surfaces from the polyhedron. This identifies and extracts the boundary surfaces of the initial 3D model, avoiding the influence of noise or uneven grayscale on the 3D reconstruction results, ensuring the integrity and accuracy of the model's structural information and positioning, and improving the model's accuracy.

[0015] Furthermore, based on the preset model simplification algorithm and the triangular mesh model, the target 3D model is obtained, including: obtaining the number of triangular faces based on the triangular mesh model; traversing the triangular faces of the triangular mesh model, if any edge of any triangular face does not have an adjacent connecting edge, then setting a skip mark for the triangular face; performing model simplification steps on all triangular faces without skip marks: eliminating triangular faces based on the preset model simplification algorithm until the number of remaining triangular faces meets the preset simplification target, and ending the model simplification steps; traversing the remaining triangular faces, if any corner point is deleted, then adding the corresponding corner point and performing model simplification steps until the remaining triangular faces meet the preset simplification target, thus obtaining the target 3D model.

[0016] In the above scheme, a topology protection mechanism is introduced before simplification of the triangular mesh model. Adjacent edge connectivity analysis is performed on each triangular face, and then redundant triangular faces are removed according to the model simplification steps. This simplifies the model while maintaining its original connectivity and closure. Furthermore, corner points are checked after simplification to prevent gaps in the model caused by deleted corner points, ensuring the integrity of the model boundaries and resulting in a more accurate 3D target model. Thus, by reducing the amount of model data while maximizing the preservation of key anatomical features, the structural information and structural positioning of the model are ensured to be complete and accurate, thereby improving the model's accuracy.

[0017] Furthermore, the physical spatial center is obtained based on the pre-acquired oral CBCT data, and the image center of the target 3D model is registered with the physical spatial center based on a preset model spatial registration algorithm. This includes: obtaining the physical spatial center based on the spatial structure information corresponding to the pre-acquired oral CBCT data; obtaining the center of the target 3D model and mapping the physical spatial center and the center of the target 3D model to a preset spatial coordinate system to obtain the coordinates of the physical spatial center and the current spatial center coordinates of the target 3D model; subtracting the coordinates of the physical spatial center and the current spatial center coordinates of the target 3D model based on the preset model spatial registration algorithm to obtain a translation vector; and performing a translation operation on the target 3D model according to the translation vector to align the center of the target 3D model to the physical spatial center.

[0018] In the above scheme, the center of the target 3D model and the physical spatial center obtained from the pre-acquired oral CBCT data are mapped to a spatial coordinate system. Then, the difference between the physical spatial center of the image and the center of the model is calculated to obtain a translation vector. The translation operation is then performed on the target 3D model to complete the spatial registration of the target 3D model. This aligns the center of the target 3D model with the actual physical spatial center of the image in space, ensuring the integrity and accuracy of the model's structural information and structural positioning, thus improving the model's accuracy.

[0019] This invention also provides a model registration system based on three-dimensional reconstruction of oral images, comprising: an initial three-dimensional model construction module, a triangular mesh model construction module, a target three-dimensional model construction module, and a model registration module; the initial three-dimensional model construction module is used to obtain an initial three-dimensional model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data; the triangular mesh model construction module is used to obtain a triangular mesh model based on a preset three-dimensional reconstruction algorithm and the initial three-dimensional model; the target three-dimensional model construction module is used to obtain a target three-dimensional model based on a preset model simplification algorithm and the triangular mesh model; the model registration module is used to obtain the physical spatial center based on the pre-acquired oral CBCT data, and register the image center and physical spatial center of the target three-dimensional model based on a preset model spatial registration algorithm.

[0020] This invention proposes a model registration system based on 3D reconstruction of oral images. An initial 3D model construction module processes pre-acquired oral CBCT data using a preset AI segmentation algorithm to obtain an initial 3D model. A triangular mesh model construction module then constructs a triangular network model using the preset 3D model for 3D reconstruction. A target 3D model construction module further simplifies the reconstructed triangular network model using a preset model simplification algorithm to obtain the target 3D model, thereby improving the model's geometric accuracy. Finally, the physical spatial center is obtained from the pre-acquired oral CBCT data, and the model registration module registers the physical spatial center with the image center of the target 3D model according to a preset model spatial registration algorithm, ensuring alignment between the model and the original image in the physical coordinate system. Thus, by employing steps such as AI segmentation, 3D reconstruction, model simplification, and spatial registration to simplify and spatially register the 3D model, the system ensures the integrity and accuracy of the model's structural information and positioning, thereby improving the model's precision.

[0021] Furthermore, the initial 3D model construction module is used to obtain an initial 3D model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data. It includes: a data acquisition unit, a path construction unit, a spatial structure information acquisition unit, and a model construction unit. The data acquisition unit is used to acquire several slice files based on the pre-acquired oral CBCT data. The path construction unit is used to sort the slice files based on their file numbers to obtain slice file paths. The spatial structure information acquisition unit is used to traverse each slice file in the slice file path to obtain the number of slices and the physical information of each slice file, and to obtain spatial structure information based on the number of slices and the physical information of each slice file. The model construction unit is used to obtain the initial 3D model based on the preset AI segmentation algorithm and the spatial structure information.

[0022] In the above scheme, the data acquisition unit and path construction unit sort the slice files corresponding to the pre-acquired oral CBCT data, and construct the slice file paths based on the sorting results. This ensures that the pre-acquired oral CBCT data maintains the correct spatial order and structural continuity during the 3D reconstruction process. The spatial structure information acquisition unit then extracts the number of slices and physical information to construct accurate spatial structure information to provide an accurate geometric basis. Finally, the model construction unit combines a preset AI segmentation algorithm to construct an initial 3D model. Therefore, accurate analysis of the pre-acquired oral CBCT data provides an accurate data foundation for subsequent 3D reconstruction, model simplification, and spatial registration steps, ensuring the integrity and accuracy of the model's structural information and structural positioning, and improving the model's accuracy.

[0023] Furthermore, the model registration module is used to obtain the physical spatial center based on the pre-acquired oral CBCT data, and to register the image center of the target 3D model with the physical spatial center based on a preset model spatial registration algorithm. This module includes: a physical spatial center acquisition unit, a spatial coordinate mapping unit, a translation vector calculation unit, and a translation registration unit. The physical spatial center acquisition unit is used to obtain the physical spatial center based on the spatial structure information corresponding to the pre-acquired oral CBCT data. The spatial coordinate mapping unit is used to obtain the center of the target 3D model and map the physical spatial center and the center of the target 3D model to a preset spatial coordinate system to obtain the coordinates of the physical spatial center and the current spatial center coordinates of the target 3D model. The translation vector calculation unit is used to calculate the difference between the coordinates of the physical spatial center and the current spatial center coordinates of the target 3D model based on the preset model spatial registration algorithm to obtain a translation vector. The translation registration unit is used to perform a translation operation on the target 3D model according to the translation vector, aligning the center of the target 3D model to the physical spatial center.

[0024] In the above scheme, the physical space center acquisition unit and the spatial coordinate mapping unit map the center of the target 3D model and the physical space center obtained from the pre-acquired oral CBCT data to the spatial coordinate system. The translation vector calculation unit then calculates the difference between the physical space center of the image and the center of the model to obtain the translation vector. The translation registration unit performs a translation operation on the target 3D model to complete the spatial registration of the target 3D model. Thus, the center of the target 3D model is spatially aligned with the actual physical space center of the image, ensuring the integrity and accuracy of the model's structural information and structural positioning, thereby improving the model's accuracy. Attached Figure Description

[0025] Figure 1 A schematic flowchart illustrating the steps of a model registration method based on three-dimensional reconstruction of oral images, provided in a certain embodiment of the present invention;

[0026] Figure 2A schematic diagram of the execution flow of a preset AI segmentation algorithm for a model registration method based on three-dimensional reconstruction of oral images, provided in a certain embodiment of the present invention;

[0027] Figure 3 A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 1 ;

[0028] Figure 4 A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 2 ;

[0029] Figure 5 A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 3 ;

[0030] Figure 6 A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 4 ;

[0031] Figure 7 This is a schematic diagram of the module structure of a model registration system based on three-dimensional reconstruction of oral images, provided in a certain embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Example 1

[0034] See Figure 1 , Figure 1 This is a flowchart illustrating the steps of a model registration method based on three-dimensional reconstruction of oral images, provided in a certain embodiment of the present invention; as shown below. Figure 1 As shown in the figure, this embodiment of the invention proposes a model registration method based on three-dimensional reconstruction of oral images, including steps 101 to 104, each step of which is as follows:

[0035] Step 101: Based on the preset AI segmentation algorithm and the pre-acquired oral CBCT data, an initial three-dimensional model is obtained;

[0036] Step 102: Based on the preset 3D reconstruction algorithm and the initial 3D model, a triangular mesh model is obtained;

[0037] Step 103: Based on the preset model simplification algorithm and triangular mesh model, obtain the target 3D model;

[0038] Step 104: Obtain the physical spatial center based on the pre-acquired oral CBCT data, and register the image center and physical spatial center of the target 3D model based on the preset model spatial registration algorithm.

[0039] One possible implementation involves extracting necessary spatial structure information from the raw image data. This spatial structure information includes resolution and voxel dimensions. In this embodiment, the raw image data can be oral CBCT data, and the input format for oral CBCT data can be DICOM (.dcm). Alternatively, the raw image data can also be conventional CT files and magnetic resonance imaging (MRI) files, with supported input formats including NIfTI (.nii.gz) and MetaImage (.mha, .mhd). After obtaining the oral CBCT data, a preset AI segmentation algorithm is used to extract the three-dimensional structure of the oral CBCT data. In this implementation, the preset AI segmentation algorithm can use Unet based on an encoder-decoder structure and its variants, such as 3D Unet, Residual Unet, and Attention Unet. It can also use various network models based on the Transformer architecture, such as TransUNet and Swin-UNETR. The three-dimensional model generated by the preset AI segmentation algorithm is then converted into a data format supported by a visualization tool (Visualization Toolkit, VTK) to obtain the initial three-dimensional model. The initial 3D model obtained from segmentation is then visualized and analyzed using a preset 3D reconstruction algorithm, converting it into a visualized triangular mesh model. In this embodiment, the preset 3D reconstruction algorithm can be the Marching Cubes algorithm, a classic 3D surface reconstruction algorithm widely used in medical image processing, 3D modeling, and visualization analysis to achieve 3D reconstruction of target structures. To improve the efficiency of the triangular mesh model in visualization and post-processing, this embodiment also proposes to simplify the triangular mesh model by reducing the number of triangular faces. This improves the reading and processing efficiency of the triangular mesh model without losing surface information. In addition, after simplification and compression, the model surface often has jagged edges, burrs, or local unevenness. To avoid affecting the use of the triangular mesh model in subsequent visualization processes, this embodiment also proposes to smooth the surface of the triangular mesh model. This is achieved by optimizing the model using a preset surface smoothing algorithm. In this embodiment, the preset surface smoothing algorithm can be an iterative smoothing filter, thereby obtaining a smooth, continuous, and aesthetically pleasing target 3D model.Finally, the position of the target 3D model in physical space is aligned with the physical space center corresponding to the pre-acquired CBCT data, so that the center of the target 3D model coincides with the physical space center. This completes the accurate registration of the model in the coordinate system, giving the target 3D model an accurate positional reference to other influences during visualization analysis. It is worth mentioning that after completing the step of aligning the position of the target 3D model in physical space with the physical space center corresponding to the pre-acquired CBCT data, the allclose function of the NumPy library can generally be called to check whether the center point of the registered target 3D model meets the preset accuracy requirements of the physical space center, in order to ensure the accuracy of the model. The preset accuracy requirements are usually set according to the actual application, for example, the preset accuracy requirement is 1×10. -5 Millimeters.

[0040] This invention proposes a model registration method based on 3D reconstruction of oral images. After processing pre-acquired oral CBCT data using a preset AI segmentation algorithm to obtain an initial 3D model, a triangular network model is first constructed using the preset 3D model to perform 3D reconstruction. Then, the reconstructed triangular network model is simplified using a preset model simplification algorithm to obtain the target 3D model, thereby improving the geometric accuracy of the model. Finally, the physical spatial center is obtained from the pre-acquired oral CBCT data, and the physical spatial center is registered with the image center of the target 3D model according to a preset model spatial registration algorithm, ensuring that the model and the original image are aligned in the physical coordinate system. Thus, by employing steps such as AI segmentation, 3D reconstruction, model simplification, and spatial registration to simplify and spatially register the 3D model, the structural information and structural positioning of the model are ensured to be complete and accurate, improving the model's precision.

[0041] A preferred approach involves obtaining an initial 3D model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data, including: acquiring several slice files based on the pre-acquired oral CBCT data; sorting the slice files according to their file numbers to obtain slice file paths; traversing each slice file in the slice file path to obtain the number of slices and the physical information of each slice file, and obtaining spatial structure information based on the number of slices and the physical information of each slice file; and obtaining the initial 3D model based on the preset AI segmentation algorithm and the spatial structure information.

[0042] One preferred implementation involves extracting necessary spatial structure information from the raw image data. This spatial structure information includes resolution and voxel dimensions. In this embodiment, the raw image data can be oral CBCT data, with the input format being DICOM (.dcm). Alternatively, the raw image data can be conventional CT files and magnetic resonance imaging (MRI) files. Supported input formats include NIfTI (.nii.gz) and MetaImage (.mha, .mhd). Specifically, taking oral CBCT data in DICOM format as an example, each slice of the DICOM format oral CBCT data is treated as a separate file, represented as a slice file. In oral CBCT data, DICOM slices may not be arranged in a fixed order; therefore, they must be sorted according to information such as the file number (InstanceNumbe) within the slice file to ensure the correct continuity of the anatomical structure after 3D reconstruction. After sorting, the os.listdir() and os.path.join() functions from the Python function library are called to obtain the filenames of all slice files and construct the complete slice file path. Next, the physical information of each slice file is obtained from the slice file path. One interpretation is to call the dcmread() function from the Python function library to read any DICOM image, and extract the actual size represented by each voxel in three-dimensional space through its pixelSpacing information and sliceThickness information. The pixelSpacing information represents the physical pixel spacing along the row direction and the physical pixel spacing along the column direction of the image matrix, and the sliceThickness information represents the thickness between adjacent slices, that is, the physical size of the voxel in the axial direction. Then, the row information and column information of the image are extracted to represent the two-dimensional size of a single slice. The row information represents the number of rows of the image in the two-dimensional plane, and the column information represents the number of columns of the image in the two-dimensional plane. The physical information and the number of slices are combined to represent the spatial structure information of each slice. That is, the voxel dimension is formed by the number of rows, columns and slices read, which represents the number of discrete grids of volume data in three axes. Combined with the resolution of the DICOM image, the size of each voxel in real physical space is represented, that is, the "physical size of the voxel". By combining the voxel dimension and resolution, the actual range covered by the DICOM image in three-dimensional space can be derived. Finally, the initial three-dimensional model is constructed by using a preset AI segmentation algorithm and spatial structure information.

[0043] In the above scheme, the slice files corresponding to the pre-acquired oral CBCT data are sorted, and the slice file paths are constructed based on the sorting results. This ensures that the pre-acquired oral CBCT data maintains the correct spatial order and structural continuity during 3D reconstruction. Then, the number of slices and physical information are extracted to construct accurate spatial structural information, providing an accurate geometric basis. Finally, an initial 3D model is constructed using a preset AI segmentation algorithm. Therefore, accurate analysis of the pre-acquired oral CBCT data provides an accurate data foundation for subsequent 3D reconstruction, model simplification, and spatial registration steps, ensuring the integrity and accuracy of the model's structural information and location, and improving the model's precision.

[0044] A preferred approach, based on a preset AI segmentation algorithm and spatial structure information, obtains an initial 3D model, comprising: automatically segmenting pre-acquired oral CBCT data using a preset AI segmentation algorithm to obtain a first oral mask image; traversing several voxels in the first oral mask image, identifying corner voxels and setting the corner values ​​as preset foreground values ​​to obtain a second oral mask image; constructing the second oral mask image into a visual image object, and based on the spatial structure information, setting the spatial dimension and actual physical size of the visual image object to obtain the initial 3D model.

[0045] One preferred implementation involves acquiring oral CBCT data and then using a pre-defined AI segmentation algorithm to extract the 3D structure of the oral CBCT data. In this implementation, the pre-defined AI segmentation algorithm can employ UNet and its variants based on an encoder-decoder structure, such as 3D UNet, Residual UNet, and Attention UNet. It can also employ various network models based on the Transformer architecture, such as TransUNet and Swin-UNETR. The 3D model generated by the pre-defined AI segmentation algorithm is then converted into a data format supported by a visualization tool (Visualization Toolkit, VTK) to obtain the initial 3D model. See details below. Figure 2 , Figure 2 This is a schematic diagram illustrating the execution flow of a preset AI segmentation algorithm for a model registration method based on three-dimensional reconstruction of oral images, provided in a certain embodiment of the present invention; as shown below. Figure 2 As shown, the existing mature Unet algorithm based on an encoder-decoder structure is used to automatically segment the pre-acquired oral CBCT data. First, the pre-acquired oral CBCT data is read, and then the pre-processed oral CBCT data is input into the Unet model. Automatic segmentation is performed in the pre-trained Unet model, and finally, the jawbone structure region is segmented and extracted from the oral CBCT data. The specific algorithm implementation has mature technologies and will not be explained in detail here; see [link to documentation]. Figure 3 , Figure 3A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 1 ;like Figure 3 As shown, after automatic segmentation based on the preset AI segmentation algorithm, the ReadImage() and GetArrayFromImage() functions are called from the SimpleITK library to output the segmentation result as a mask image, representing the first oral cavity mask image. Then, the voxels in the mask image are traversed, and corner points are added to the mask image. The corner point value corresponding to the voxel at the corner point position is set as the preset foreground value. In this embodiment, the preset foreground value is set to 1. At this time, the mask image represents the second oral cavity mask image. The expression of each corner point is as follows: (x_min, y_min, z_min), (x_max, y_min, z_min), (x_min, y_max, z_min), (x_min, y_min, z_max), (x_max, y_max, z_min ... The functions z_min, (x_max, y_min, z_max), (x_min, y_max, z_max), and (x_max, y_max, z_max) are used. Finally, the vtk.vtkImageData() function is called to create a VTK image object, i.e., a visualization image object, from the second oral mask image. Based on the physical information of each slice file, the SetDimensions() function is called to set the number of voxels in the three axes of the visualization image object, i.e., the spatial dimensions of the image, and the SetSpacing() function is called to set the size of each voxel in the actual physical space. The unit of this size is usually millimeters. Finally, an initial three-dimensional model of the visualization image object is obtained, which maintains the same physical size as the original oral CBCT data in physical space.

[0046] In the above scheme, based on the AI ​​segmentation algorithm, the pre-acquired oral CBCT data is automatically segmented to obtain a first oral mask image. Then, foreground values ​​are assigned to the corner voxels of the first oral mask image to obtain a second oral mask image. The processed second oral mask image is converted into a visual image object with spatial dimensions and actual physical space size, ensuring that the initial 3D model maintains structural integrity. Thus, the 2D image is converted into a 3D model. Through mask assignment, not only is the structural integrity of the 3D model guaranteed, but the 3D model also has physical spatial attributes consistent with the actual image, ensuring the integrity and accuracy of the model's structural information and structural positioning, thereby improving the model's accuracy.

[0047] A preferred embodiment, based on a preset 3D reconstruction algorithm and an initial 3D model, obtains a triangular mesh model, comprising: constructing adjacent voxels in the initial 3D model into several polyhedral elements based on the preset 3D reconstruction algorithm; extracting boundary surfaces from the several polyhedra based on preset isosurfaces; traversing the several polyhedra and generating triangular patches based on the boundary surfaces; and meshing the triangular patches to obtain a triangular mesh model.

[0048] One preferred implementation involves using a preset 3D reconstruction algorithm to perform visual analysis on the segmented initial 3D model, converting it into a visualized triangular mesh model. In this embodiment, the preset 3D reconstruction algorithm can be the Marching Cubes algorithm. The Marching Cubes algorithm is a classic 3D surface reconstruction algorithm widely used in medical image processing, 3D modeling, and visualization analysis to achieve 3D reconstruction of target structures. For details, see... Figure 4 , Figure 4 A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 2 ;like Figure 4 As shown, the initial 3D model is first treated as a series of adjacent small polyhedra using the Marching Cubes algorithm. In this embodiment, a cube is used as an example. The vtkMarchingCubes class from the VTK library is called to process the initial 3D model. By setting the preset isosurface to 0.5, boundary surfaces are extracted from the cubes. Then, the Update() function is called to execute the Marching Cubes algorithm, which traverses all local cubes in the entire 3D data, automatically completing boundary judgment, table lookup, generation of triangular facets, and mesh stitching of the triangular facets. The output is a closed triangular mesh model. Taking a voxel as a cube as an example, based on the combination of the current cube's 8 vertex states, the Marching Cubes algorithm uses a built-in preset lookup table to directly determine which triangular facets should be generated inside the current cube to approximate the surface of the current cube, thereby quickly constructing a continuous 3D mesh and obtaining a triangular mesh model. A schematic diagram of the specific triangular mesh model is shown below. Figure 4 As shown. It's worth noting that the obtained triangular mesh model can be exported as a standard 3D model file using the VTK write interface. The VTK write interface includes the vtkSTLWriter() or vtkOBJWriter() functions; standard 3D model files include STL files and OBJ files, etc.

[0049] In the above scheme, voxels in the initial 3D model are constructed into polyhedral units, and boundary surfaces are extracted based on preset isosurfaces. Triangular patches are then generated from these boundary surfaces, and after mesh stitching, a triangular mesh model is obtained. This allows a preset 3D reconstruction algorithm to accurately capture the contours of the complex anatomical structures of the initial 3D model, generating a continuous and closed triangular mesh model. Therefore, by converting volume data into a triangular mesh, the accuracy of 3D reconstruction is ensured while also improving visualization effects. This guarantees the integrity and accuracy of the model's structural information and positioning, thus enhancing the model's precision.

[0050] A preferred embodiment involves extracting boundary surfaces from several polyhedra based on a preset isosurface, comprising: mapping several voxels in an initial 3D model to binary masks; dividing voxels that do not meet the preset mask requirements into a first voxel group according to the preset isosurface; dividing voxels that meet the preset mask requirements into a second voxel group; and extracting boundary surfaces from several polyhedra based on the first voxel group and the second voxel group.

[0051] One preferred implementation involves a large number of voxels in the initial 3D model. Each voxel is assigned a corresponding label using a binary mask, where 0 represents the background and 1 represents the target region. The Marching Cubes algorithm treats the initial 3D model as composed of several adjacent cubes, each cube having 8 adjacent voxels whose vertices are defined by their values. For each cube, the Marching Cubes algorithm determines whether the cube is located in the target region based on whether the mask values ​​of these 8 vertices are higher than a preset isosurface. Since the mask value is set to binary (1 or 0), the preset isosurface is set to 0 or 5. If the voxel's mask value is greater than the preset isosurface value, the voxel is marked as outside the region, and all voxels marked as outside the region are grouped into a first voxel group. If the voxel's mask value is less than the preset isosurface value, the voxel is marked as inside the region, and all voxels marked as inside the region are grouped into a second voxel group. The preset mask requires that the voxel's mask value be less than the preset isosurface value. In this embodiment, since the initial 3D model is considered as having several small cubes, the relationship between the voxel values ​​of the vertices and the preset isosurface is directly determined to determine the position of the cube. Then, the extraction boundary surface is extracted from the cube based on the first and second voxel groups.

[0052] In the above scheme, voxels are mapped to binary masks, and voxels are grouped according to isosurfaces. The target region is distinguished from the background region by the first and second voxel groups to extract the boundary surfaces from the polyhedron. This identifies and extracts the boundary surfaces of the initial 3D model, avoiding the influence of noise or uneven grayscale on the 3D reconstruction results, ensuring the integrity and accuracy of the model's structural information and positioning, and improving the model's accuracy.

[0053] A preferred scheme for obtaining a target 3D model based on a preset model simplification algorithm and a triangular mesh model includes: obtaining the number of triangular faces based on the triangular mesh model; traversing the triangular faces of the triangular mesh model, and if any edge of any triangular face does not have an adjacent connecting edge, setting a skip mark for the triangular face; performing a model simplification step on all triangular faces without a skip mark: eliminating triangular faces based on the preset model simplification algorithm until the number of remaining triangular faces meets the preset simplification target, and ending the model simplification step; traversing the remaining triangular faces, and if any corner point is deleted, adding the corresponding corner point and performing the model simplification step again until the remaining triangular faces meet the preset simplification target, thus obtaining the target 3D model.

[0054] In a preferred implementation, to improve the efficiency of triangular mesh models in visualization and post-processing, this embodiment of the invention proposes to simplify the triangular mesh model. By reducing the number of triangular faces in the triangular mesh model, the reading and processing efficiency of the triangular mesh model is improved without losing surface information. Specifically, during the simplification process, a preset model simplification algorithm prioritizes simplifying areas with relatively smooth structural surfaces and minimal shape changes in the triangular mesh model, such as relatively flat parts of the model surface; while retaining as much as possible areas with edges, protrusions, or large curvature, thereby reducing the impact on the overall structure recognition. After each simplification step, the system automatically adjusts the connections of surrounding points and faces to ensure a smooth transition and geometric continuity of the model shape. In this embodiment, the preset model simplification algorithm uses the SetTargetReduction() function to delete a preset proportion of triangular faces. Based on the preset simplification target, triangular faces in the 3D model are selectively deleted. The preset simplification target is generally set according to actual needs and is not limited here. One explanation is to first count the number of original triangles in the triangular mesh model, then preset the simplification target (e.g., 95%), calculate the final number of faces to be retained, and then use `SetTargetReduction(0.95)` to delete 95% of the triangles while preserving the basic shape, leaving only 5%. For a comparison before and after simplification, please refer to [link to previous explanation]. Figure 5 , Figure 5 A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 3 ,like Figure 5As shown, simplifying the number of triangles does not cause loss of surface information, but it speeds up model reading and file processing. It's worth noting that before each deletion operation, the `PreserveTopologyOn()` function is called to prevent holes, structural breaks, and other issues from occurring during simplification, ensuring the connectivity and recognizability of the anatomical structure. Specifically, before each deletion operation, the entire triangular mesh model is checked to determine if the deletion will disrupt its overall connectivity or closure. If a triangle is the only connection to an edge, deleting it might create a "gap," and the operation will be automatically skipped. After each deletion operation, it's necessary to check if the eight corner points of the triangular mesh model have been deleted. If they were deleted due to simplification, they are added back, and the deletion operation is repeated until the preset simplification goal is met, resulting in the target 3D model. The preset simplification goal can be set according to the actual application.

[0055] As an example of an embodiment of the present invention, after the triangular mesh model is simplified and compressed, the model surface will usually have problems such as jagged edges, burrs or local unevenness. In order not to affect the use effect of the triangular mesh model in the subsequent visualization process, the present invention also proposes to smooth the surface of the triangular mesh model. The model is optimized by using a preset surface smoothing algorithm. In this embodiment, the preset surface smoothing algorithm can use an iterative smoothing filter to obtain a smooth, continuous and beautiful target 3D model. Specifically, the simplified triangular mesh undergoes multiple rounds of local position adjustments using an iterative smoothing filter. In each iteration, a new position is calculated based on the positional relationship between each point and its adjacent vertices to reduce surface unevenness, thereby gradually smoothing the model's appearance. The relaxation factor controls the magnitude of each point's movement; a smaller value results in a gentler change, while a larger value accelerates the smoothing process. To prevent deformation or structural loss at the model boundaries due to smoothing, a boundary protection mechanism is enabled, ensuring that edge points are constrained by the original contour lines during position updates, thus preserving the overall structural features of the model. After smoothing, an optimized mesh model is generated through an update process and exported in a standard 3D file format. See the export results for details. Figure 6 , Figure 6 A schematic diagram of a three-dimensional model for a model registration method based on three-dimensional reconstruction of oral images provided in a certain embodiment of the present invention. Figure 4 ;like Figure 6As shown, the overall structural features of the target 3D model are preserved, and the surface unevenness is reduced, resulting in a smooth model appearance. One specific explanation involves setting a smoothing filter and using the vtkSmoothPolyDataFilter() function to iteratively smooth the simplified mesh. The number of smoothing iterations and the relaxation factor are set; in this embodiment, the number of smoothing iterations can be set to 50, and the relaxation factor can be set to 0.3–0.5. The BoundarySmoothingOn() function is called to protect the model boundaries, ensuring that the edge contours transition smoothly while preserving details, preventing over-smoothing from causing the loss of model contours.

[0056] In the above scheme, a topology protection mechanism is introduced before simplification of the triangular mesh model. Adjacent edge connectivity analysis is performed on each triangular face, and then redundant triangular faces are removed according to the model simplification steps. This simplifies the model while maintaining its original connectivity and closure. Furthermore, corner points are checked after simplification to prevent gaps in the model caused by deleted corner points, ensuring the integrity of the model boundaries and resulting in a more accurate 3D target model. Thus, by reducing the amount of model data while maximizing the preservation of key anatomical features, the structural information and structural positioning of the model are ensured to be complete and accurate, thereby improving the model's accuracy.

[0057] A preferred embodiment involves obtaining the physical spatial center based on pre-acquired oral CBCT data, and registering the image center of the target 3D model with the physical spatial center based on a preset model spatial registration algorithm. This includes: obtaining the physical spatial center based on the spatial structure information corresponding to the pre-acquired oral CBCT data; obtaining the center of the target 3D model and mapping the physical spatial center and the center of the target 3D model to a preset spatial coordinate system to obtain the coordinates of the physical spatial center and the current spatial center coordinates of the target 3D model; subtracting the physical spatial center coordinates from the current spatial center coordinates of the target 3D model using the preset model spatial registration algorithm to obtain a translation vector; and performing a translation operation on the target 3D model according to the translation vector to align the center of the target 3D model to the physical spatial center.

[0058] One preferred implementation involves aligning the position of the target 3D model in physical space with the physical space center corresponding to the pre-acquired CBCT data. This ensures the center of the target 3D model coincides with the physical space center, achieving precise registration of the model in the coordinate system and providing accurate positional references for the target 3D model when used in visualization analysis. Specifically, to ensure the generated 3D mesh model's position in physical space remains consistent with the original medical image, precise alignment of spatial coordinates is required. First, the coordinates of the physical space center point of the original oral CBCT image are acquired. The physical space center point coordinates refer to the center point of the image in 3D physical space. The center point is mapped to the spatial coordinate system using the following formula:

[0059] Physical space center point coordinates = [number of rows × pixel spacing / 2, number of columns × pixel spacing / 2, number of slices × layer thickness / 2];

[0060] Load the target 3D model using the trimesh library, and map the target 3D model to a spatial coordinate system using the bounding_box.centroid() or center_mass() function to obtain its center coordinates, which serve as the current spatial center of the target 3D model. The current spatial center represents the position of the center point of the target 3D model in the 3D coordinate system, which can usually be obtained by calculating the geometric center or mass center of the model's bounding box. The coordinate difference between the current spatial center coordinates and the physical spatial center coordinates of the target 3D model is then calculated as a translation vector, representing the required spatial offset of the model. By constructing a translation matrix and applying it to the entire model point set, the target 3D model can be translated as a whole, aligning its current spatial center with the physical spatial center, thus achieving coordinate alignment. Specifically, the preset model spatial registration algorithm uses the `translation_matrix()` and `apply_translation()` functions of trimesh to perform translation on the model, aligning the target 3D model's current spatial center to the physical spatial center. After aligning the target 3D model's position in physical space with the physical spatial center corresponding to the pre-acquired CBCT data, ensuring the center of the target 3D model coincides with the physical spatial center, the `allclose` function of the NumPy library can generally be called to check whether the center point of the registered target 3D model meets the preset accuracy requirements of the physical spatial center, ensuring model accuracy. The preset accuracy requirements are usually set based on the actual application, for example, a preset accuracy requirement of 1×10⁻⁶. -5 Millimeters.

[0061] In the above scheme, the center of the target 3D model and the physical spatial center obtained from the pre-acquired oral CBCT data are mapped to a spatial coordinate system. Then, the difference between the physical spatial center of the image and the center of the model is calculated to obtain a translation vector. The translation operation is then performed on the target 3D model to complete the spatial registration of the target 3D model. This aligns the center of the target 3D model with the actual physical spatial center of the image in space, ensuring the integrity and accuracy of the model's structural information and structural positioning, thus improving the model's accuracy.

[0062] Example 2

[0063] See Figure 7 , Figure 7 This is a schematic diagram of the module structure of a model registration system based on three-dimensional reconstruction of oral images, provided in one embodiment of the present invention. Figure 7 As shown, this embodiment of the invention also provides a model registration system based on three-dimensional reconstruction of oral images, including: an initial three-dimensional model construction module 201, a triangular mesh model construction module 202, a target three-dimensional model construction module 203, and a model registration module 204; the initial three-dimensional model construction module 201 is used to obtain an initial three-dimensional model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data; the triangular mesh model construction module 202 is used to obtain a triangular mesh model based on a preset three-dimensional reconstruction algorithm and the initial three-dimensional model; the target three-dimensional model construction module 203 is used to obtain a target three-dimensional model based on a preset model simplification algorithm and the triangular mesh model; the model registration module 204 is used to obtain the physical space center based on the pre-acquired oral CBCT data, and register the image center and physical space center of the target three-dimensional model based on a preset model space registration algorithm.

[0064] This invention proposes a model registration system based on 3D reconstruction of oral images. An initial 3D model construction module processes pre-acquired oral CBCT data using a preset AI segmentation algorithm to obtain an initial 3D model. A triangular mesh model construction module then constructs a triangular network model using the preset 3D model for 3D reconstruction. A target 3D model construction module further simplifies the reconstructed triangular network model using a preset model simplification algorithm to obtain the target 3D model, thereby improving the model's geometric accuracy. Finally, the physical spatial center is obtained from the pre-acquired oral CBCT data, and the model registration module registers the physical spatial center with the image center of the target 3D model according to a preset model spatial registration algorithm, ensuring alignment between the model and the original image in the physical coordinate system. Thus, by employing steps such as AI segmentation, 3D reconstruction, model simplification, and spatial registration to simplify and spatially register the 3D model, the system ensures the integrity and accuracy of the model's structural information and positioning, thereby improving the model's precision.

[0065] In a preferred embodiment, the initial 3D model construction module 201 is used to obtain an initial 3D model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data, including: a data acquisition unit 301, a path construction unit 302, a spatial structure information acquisition unit 303, and a model construction unit 304; the data acquisition unit 301 is used to acquire several slice files based on the pre-acquired oral CBCT data; the path construction unit 302 is used to sort several slice files based on the file numbers in the several slice files to obtain slice file paths; the spatial structure information acquisition unit 303 is used to traverse each slice file in the slice file path to obtain the number of slices and the physical information of each slice file, and to obtain spatial structure information based on the number of slices and the physical information of each slice file; the model construction unit 304 is used to obtain the initial 3D model based on the preset AI segmentation algorithm and the spatial structure information.

[0066] In the above scheme, the data acquisition unit and path construction unit sort the slice files corresponding to the pre-acquired oral CBCT data, and construct the slice file paths based on the sorting results. This ensures that the pre-acquired oral CBCT data maintains the correct spatial order and structural continuity during the 3D reconstruction process. The spatial structure information acquisition unit then extracts the number of slices and physical information to construct accurate spatial structure information to provide an accurate geometric basis. Finally, the model construction unit combines a preset AI segmentation algorithm to construct an initial 3D model. Therefore, accurate analysis of the pre-acquired oral CBCT data provides an accurate data foundation for subsequent 3D reconstruction, model simplification, and spatial registration steps, ensuring the integrity and accuracy of the model's structural information and structural positioning, and improving the model's accuracy.

[0067] In a preferred embodiment, the model registration module 204 is used to obtain the physical spatial center based on the pre-acquired oral CBCT data, and to register the image center of the target 3D model with the physical spatial center based on a preset model spatial registration algorithm. The module includes: a physical spatial center acquisition unit 401, a spatial coordinate mapping unit 402, a translation vector calculation unit 403, and a translation registration unit 404. The physical spatial center acquisition unit 401 is used to obtain the physical spatial center based on the spatial structure information corresponding to the pre-acquired oral CBCT data. The spatial coordinate mapping unit 402 is used to obtain the center of the target 3D model and map the physical spatial center and the center of the target 3D model to a preset spatial coordinate system to obtain the coordinates of the physical spatial center and the current spatial center coordinates of the target 3D model. The translation vector calculation unit 403 is used to calculate the difference between the coordinates of the physical spatial center and the current spatial center coordinates of the target 3D model based on the preset model spatial registration algorithm to obtain a translation vector. The translation registration unit 404 is used to perform a translation operation on the target 3D model according to the translation vector, aligning the center of the target 3D model to the physical spatial center.

[0068] In the above scheme, the physical space center acquisition unit and the spatial coordinate mapping unit map the center of the target 3D model and the physical space center obtained from the pre-acquired oral CBCT data to the spatial coordinate system. The translation vector calculation unit then calculates the difference between the physical space center of the image and the center of the model to obtain the translation vector. The translation registration unit performs a translation operation on the target 3D model to complete the spatial registration of the target 3D model. Thus, the center of the target 3D model is spatially aligned with the actual physical space center of the image, ensuring the integrity and accuracy of the model's structural information and structural positioning, thereby improving the model's accuracy.

[0069] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0070] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the described specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0071] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

Claims

1. A model registration method based on three-dimensional reconstruction of oral images, characterized in that, include: An initial three-dimensional model is obtained based on a preset AI segmentation algorithm and pre-acquired oral CBCT data; Based on the preset 3D reconstruction algorithm and the initial 3D model, a triangular mesh model is obtained; Based on the preset model simplification algorithm and the triangular mesh model, a target 3D model is obtained, including: obtaining the number of triangular faces based on the triangular mesh model; traversing the triangular faces of the triangular mesh model, if any edge of any triangular face does not have an adjacent connecting edge, then setting a skip mark for the triangular face; performing a model simplification step on all triangular faces without a skip mark: removing the triangular faces based on the preset model simplification algorithm until the number of remaining triangular faces meets the preset simplification target, and ending the model simplification step; traversing the remaining triangular faces, if any corner point is deleted, then adding the corresponding corner point and performing the model simplification step until the remaining triangular faces meet the preset simplification target, thus obtaining the target 3D model; Obtaining the physical space center based on the pre-acquired oral CBCT data, and registering the center of the target 3D model with the physical space center based on a preset model space registration algorithm includes: obtaining the physical space center based on the spatial structure information corresponding to the pre-acquired oral CBCT data; obtaining the center of the target 3D model, and mapping the physical space center and the center of the target 3D model to a preset spatial coordinate system to obtain the coordinates of the physical space center and the current spatial center coordinates of the target 3D model; subtracting the coordinates of the physical space center and the current spatial center coordinates of the target 3D model based on the preset model space registration algorithm to obtain a translation vector; and performing a translation operation on the target 3D model according to the translation vector to align the center of the target 3D model to the physical space center.

2. The model registration method based on three-dimensional reconstruction of oral images according to claim 1, wherein, The initial three-dimensional model, obtained based on a preset AI segmentation algorithm and pre-acquired oral CBCT data, includes: Based on the pre-acquired oral CBCT data, several slice files are obtained; The slice files are sorted based on their file numbers to obtain the slice file paths; The number of slices and the physical information of each slice file are obtained by traversing the slice file path, and the spatial structure information is obtained based on the number of slices and the physical information of each slice file. Based on the preset AI segmentation algorithm and the spatial structure information, an initial three-dimensional model is obtained.

3. The method of claim 2, wherein the method further comprises: Based on the preset AI segmentation algorithm and the spatial structure information, an initial 3D model is obtained, including: The pre-acquired oral CBCT data is automatically segmented using a preset AI segmentation algorithm to obtain the first oral mask image; Traverse several voxels in the first oral cavity mask image, identify corner voxels and set the corner values ​​to preset foreground values ​​to obtain the second oral cavity mask image; The second oral cavity mask image is constructed into a visual image object, and based on the spatial structure information, the spatial dimension and actual physical size of the visual image object are set to obtain an initial three-dimensional model.

4. The model registration method based on three-dimensional reconstruction of oral images according to claim 3, wherein, The process of obtaining a triangular mesh model based on a preset 3D reconstruction algorithm and the initial 3D model includes: Based on a preset 3D reconstruction algorithm, adjacent voxels in the initial 3D model are constructed into several polyhedral units; Based on a preset isosurface, the boundary surface is extracted from several polyhedra; Traverse several of the aforementioned polyhedra and generate triangular patches based on the boundary surfaces; The triangular facets are meshed together to obtain a triangular mesh model.

5. The model registration method based on three-dimensional reconstruction of oral images as described in claim 4, characterized in that, Based on a preset isosurface, boundary surfaces are extracted from several polyhedra, including: Map several voxels in the initial 3D model to binary masks; Based on the preset isosurface, voxels that do not meet the preset mask requirements are divided into the first voxel group; Voxels that meet the preset mask requirements are divided into the second voxel group; Boundary surfaces are extracted from several polyhedra based on the first voxel set and the second voxel set.

6. A model registration system based on three-dimensional reconstruction of oral images, characterized in that, Performing a model registration method based on three-dimensional reconstruction of oral images as described in any one of claims 1 to 5, comprising: The module includes an initial 3D model building module, a triangular mesh model building module, a target 3D model building module, and a model registration module. The initial 3D model construction module is used to obtain an initial 3D model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data; The triangular mesh model construction module is used to obtain a triangular mesh model based on a preset 3D reconstruction algorithm and the initial 3D model; The target 3D model construction module is used to obtain a target 3D model based on a preset model simplification algorithm and the triangular mesh model. The target 3D model construction module includes: obtaining the number of triangular faces based on the triangular mesh model; traversing the triangular faces of the triangular mesh model, if any edge of any triangular face does not have an adjacent connecting edge, then setting a skip mark for the triangular face; performing a model simplification step on all triangular faces without a skip mark: removing the triangular faces based on the preset model simplification algorithm until the number of remaining triangular faces meets the preset simplification target, and ending the model simplification step; traversing the remaining triangular faces, if any corner point is deleted, then adding the corresponding corner point and performing the model simplification step until the remaining triangular faces meet the preset simplification target, thus obtaining the target 3D model; The model registration module is used to obtain the physical spatial center based on the pre-acquired oral CBCT data, and to register the center of the target 3D model with the physical spatial center based on a preset model spatial registration algorithm. The model registration module includes: a physical spatial center acquisition unit, a spatial coordinate mapping unit, a translation vector calculation unit, and a translation registration unit. The physical spatial center acquisition unit is used to obtain the physical spatial center based on the spatial structure information corresponding to the pre-acquired oral CBCT data. The spatial coordinate mapping unit is used to obtain the center of the target 3D model and map the physical spatial center and the center of the target 3D model to a preset spatial coordinate system to obtain the physical spatial center coordinates and the current spatial center coordinates of the target 3D model. The translation vector calculation unit is used to calculate the difference between the physical spatial center coordinates and the current spatial center coordinates of the target 3D model based on the preset model spatial registration algorithm to obtain a translation vector. The translation registration unit is used to perform a translation operation on the target 3D model according to the translation vector to align the center of the target 3D model to the physical spatial center.

7. The model registration system based on three-dimensional reconstruction of oral images as described in claim 6, characterized in that, The initial 3D model construction module is used to obtain an initial 3D model based on a preset AI segmentation algorithm and pre-acquired oral CBCT data, including: Data acquisition unit, path construction unit, spatial structure information acquisition unit, and model construction unit; The data acquisition unit is used to acquire several slice files based on the pre-acquired oral CBCT data; The path construction unit is used to sort the slice files based on their file numbers to obtain slice file paths; The spatial structure information acquisition unit is used to traverse each slice file in the slice file path to obtain the number of slices and the physical information of each slice file, and to obtain spatial structure information based on the number of slices and the physical information of each slice file; The model building unit is used to obtain an initial three-dimensional model based on the preset AI segmentation algorithm and the spatial structure information.

Images